By: Konrad Schwarz (no.spam.delete@this.no.spam), December 9, 2014 5:12 am
Room: Moderated Discussions
Gabriele Svelto (gabriele.svelto.delete@this.gmail.com) on December 8, 2014 7:10 pm wrote:
> Konrad Schwarz (no.spam.delete@this.no.spam) on December 7, 2014 2:28 pm wrote:
> > Atomic operations with acquire/release semantics are in the same boat, of course.
>
> No, because a mutex effectively implements full sequential consistency around the protected
> region, acquire/release is weaker. This is reflected in the implementation of both on
> POWER, for example. You can provide atomics with acquire/release semantics using only
> lwsync instructions while a mutex will need isync/sync to work properly.
Whic guarantees of a mutex are
not provided by atomics with acquire/release semantics,
such that a mutex should require more heavyweight instructions?
> Konrad Schwarz (no.spam.delete@this.no.spam) on December 7, 2014 2:28 pm wrote:
> > Atomic operations with acquire/release semantics are in the same boat, of course.
>
> No, because a mutex effectively implements full sequential consistency around the protected
> region, acquire/release is weaker. This is reflected in the implementation of both on
> POWER, for example. You can provide atomics with acquire/release semantics using only
> lwsync instructions while a mutex will need isync/sync to work properly.
Whic guarantees of a mutex are
not provided by atomics with acquire/release semantics,
such that a mutex should require more heavyweight instructions?
